OBJECTIVES:

The prevalence of Alzheimer’s disease is rapidly increasing, with an expected growth to 152.8 million cases globally in 2050. A common concern for patients and caregivers is how much time remains until progression to severe stages of dementia. Moreover, predicting time by disease stage can be beneficial for clinical decision making and policy decisions around the value and use of disease-modifying therapies (DMT). We aim to determine the sojourn time by stages of disease severity and care setting.

METHODS:

Our data included patients diagnosed with dementia in the Swedish National Register for Cognitive Disorders (SveDem). The mean age at baseline was 80 years (range 28 – 105 years), 58% female. Disease severity was captured through Mini Mental State Examination (MMSE). Institutionalization and mortality were obtained from national registries. To account for informative drop-out and immortal time bias, we imputed MMSE scores for every six months following the last available measurement until death. A multi-state model with seven states representing mild/moderate/severe disease and community/institution setting, plus death, was fitted using nested Cox survival models with covariates for age and sex

RESULTS:

The analysis included 76,747 patients (59% female; mean baseline age 81 years [range 35-105 years]). Unadjusted sojourn times in days for patients starting in mild dementia were estimated (mean, [95% confidence interval]; mild = 863 [851-875]; moderate 379 [365-393]; severe 409 [356-462]; institutionalized-mild 123 [117-129]; -moderate 201 [193-209]; -severe 397 [351-443]).

CONCLUSIONS:

This study features a novel prediction framework to predict the time spent in disease states. The model is easily extendible to pre-dementia stages and inclusion of biomarker data and is ideally suited as basis for modelling the cost effectiveness of DMT. The model is derived from a single, large data source and relies on a minimum set of assumptions, leading to very precise predictions.